Statistics

We are living in dark times, and since I have been sifting through the ashes of a career, it is no surprise that failures signal through like radioactive tracers. Below is one.

Through most of my career I have been interested in the relation between science and medicine. In truth, if what matters is what you think about in the shower, I have been more interested in the relation between science and medicine than I have been interested in either activity in isolation. If I were to use a phrase to describe my focus, although it is a term that I would not have used then, I am interested in the epistemological foundations of medical practice. Pompous, I agree. I could use another phrase: what makes medicine and doctors useful? Thinking about statistical inference is a part of this topic, but there is much more to explore.

These issues became closer to my consciousness soon after I moved to Edinburgh. My ideas about what was going on were not shared by many locally, and I was nervous about going public in person rather than in print at a Symposium hosted by the Royal College of Physicians of Edinburgh. My nervousness was well founded: whilst I liked my abstract, my talk went down badly. Not least because it was truly dreadful (and the evident failure still rankles). Jan Vandenbroucke, one of the other speakers and somebody whose work I greatly admire (his paper in the Lancet, Homoeopathy trials: Going nowhere. [Lancet.1997;350:824], was to me the most important paper published in the Lancet in the 1990s), said some kind words to me afterwards, muttering that I had tried to say far too much to an audience that was ill prepared for my speculations. All true, but he was just being kind. It was worse than that.

Anyway, some  tidying up deep in my hard drive surfaced the abstract. I still like it, but it is  a shame that at the appropriate time I was unable to explain why. 

JAMES LIND SYMPOSIUM: From scurvy to systematic reviews and clinical guidelines: how can clinical research lead to better patient care? (31-10-2003, RCPE Edinburgh)

Guidelines, Automata, Science and Algorithms

There are three great branches of science: theory, experiment, and computation. (Nick Trefethen)

Advance in the mid-third of the twentieth century, the golden age of medical research, was predicated on earlier discoveries in the nineteenth century in both physiology and medicinal chemistry (1). Genetics dominated biology in the latter third of the twentieth century and many believe changes in medical practice will owe much to genetics over the next third-century (1). I disagree, and I will give an alternative view more credence: in 30 years’ time we will look back more to Neumann and Morgenstern than we will to Watson and Crick. What the Nobel laureate Herbert Simon referred to as The Sciences of the Artificial (2), subjects which have largely been peripheral to medicine, will become central.

Over the last 20 years we have seen the first (largely inadequate, I would add) attempts to explicitly demarcate methods of obtaining and promulgating knowledge about clinical practice (3,4). This has usually taken the form of proselytising a particular set of terms – systematic reviews, evidence-based practice, guidelines and the like, terms that have little to commend them or rigour. What is interesting, however, is that they reflect a long overdue renaissance of interest with the practice of medicine and medical epistemology.

The change of emphasis from the natural to the artificial is being driven by a number of forces, mostly extraneous to biomedicine: the increasing instrumental role of science in medicine and society; the increase in corporatisation of knowledge, whether by private corporations or monopsonistic institutions like the NHS (5); the rising costs of healthcare; and a remaining inability to frame questions with broad support about how to chose between alternative disease states at the level of society (6,7).

I will try to illustrate some of these issues by the use of three examples. First, the widespread use of a mode of statistical inference largely ill-suited to medicine, namely Neyman-Pearson hypothesis testing (decision-making), and the way in which this paradigm has been used to undermine expert opinion (8). Second, I will argue that we need to think much harder about clinical practice and fashion a more appropriate theoretical underpinning for clinical behaviour. Third, I will suggest how UK medical schools, in so far as they remain interested in clinical practice, should look to alternative models, perhaps business and law schools, for ideas of how they should operate (2).

1. Rees J. Complex disease and the new clinical sciences. Science 2002; 296:698-700.
2. Simon HA. The sciences of the artificial. Cambridge, Mass. MIT Press; 1969
3. Rees J. Evidence-based medicine: the epistemology that isn’t. J Am Acad Dermatol 2000; 43:727-9.
4. Rees J. Two cultures? J Am Acad Dermatol 2002; 46:313-16.
5. Hacking I. The emergence of probability: a philosophical study of early ideas about probability, induction and statistical inference. Cambridge: Cambridge University Press; 1975.
6. Ziman J. Real science. Cambridge: Cambridge University
7. Ziman J. Non-instrumental roles of science. Sci Eng Ethics
   2003;9:17-27.
8.  Gigerenzer G, Swijtink Z, Porter T et al. The empire of chance: how probability changed science and everyday life. Cambridge: CUP; 1989.

Afterword. The symposium used structured abstracts, a habit that might have a place somewhere in this galaxy, but out of choice I would prefer to live in another one. Anyway, in the published version, it reads:

• Methods: Not submitted
• Results: Not submitted
• Conclusions: Not submitted

A fair cop.